Effective learning involves extracting key patterns of information that capture the essence of our experiences and using this information to build useful knowledge that enables predictive behavior in novel situations. Humans have multiple learning systems associated with different brain regions, yet how these learning systems interact is not well understood. Also, existing theories largely ignore how information sampling behaviors are guided by experience and goals during learning. The research presented in this proposal will use a novel theoretical perspective in combination with functional magnetic resonance imaging (fMRI) and eye tracking to investigate the mechanisms of attention and learning and their interactions during category learning. The key hypothesis is that during learning, individuals must choose what information to sample, which will be guided by the person's knowledge and current goals. A class of computational models will be developed that recasts category learning as a dynamic decision process in which attention emerges as information processing guided by the learner's goals, capacity limitations, and current knowledge. These models plan a course of action (e.g., eye movements) by looking ahead and evaluating future actions for expected profit. Experiment 1 will test the key prediction of these models that sequential sampling behavior during category learning is mediated by the interaction of attention and knowledge components. Fitting these models to eye movement and classification behavior from perfect and imperfect learners will characterize the nature of information sampling and learning processes that promote optimal category learning. In Experiments 2 and 3, model components will be linked to the specific neurobiological learning systems implicated in category learning. Experiment 2 will develop and validate a novel method of linking formal models to the learning brain by using multivariate brain patterns of fMRI data to adjudicate among competing cognitive models. The key reasoning of this method is that if a model represents the true nature of category learning, continuous measures of that model's states during learning should be reflected in trial-by-trial measures of the learning brain. Experiment 3 will employ this novel model selection method to characterize how interactions between prefrontal cortex, ventral striatum, and the medial temporal lobe support successful category learning. Understanding the neurobiological mechanisms that support attention and knowledge will provide a means of predicting and promoting effective learning. Moreover, this work has the potential to inform the development of diagnostic tools that precisely characterize cognitive impairments in clinical populations that exhibit learning deficits, such as individuals with schizophrenia, depression, and Alzheimer's disease.